[llvm] r308503 - [X86] Don't try to scale down if that exceeds the bitwidth.

[Davide Italiano via llvm-commits]

On Wed, Jul 19, 2017 at 11:09 AM, Davide Italiano via llvm-commits
<llvm-commits at lists.llvm.org> wrote:
> Author: davide
> Date: Wed Jul 19 11:09:46 2017
> New Revision: 308503
>
> URL: http://llvm.org/viewvc/llvm-project?rev=308503&view=rev
> Log:
> [X86] Don't try to scale down if that exceeds the bitwidth.
>
> Fixes the crash reported in PR33844.
>
> Added:
>     llvm/trunk/test/CodeGen/X86/pr33844.ll
> Modified:
>     llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
>
> Modified: llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp?rev=308503&r1=308502&r2=308503&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp Wed Jul 19 11:09:46 2017
> @@ -1055,7 +1055,10 @@ static bool foldMaskAndShiftToScale(Sele
>
>    // Scale the leading zero count down based on the actual size of the value.
>    // Also scale it down based on the size of the shift.
> -  MaskLZ -= (64 - X.getSimpleValueType().getSizeInBits()) + ShiftAmt;
> +  unsigned ScaleDown = (64 - X.getSimpleValueType().getSizeInBits()) + ShiftAmt;
> +  if (MaskLZ < ScaleDown)
> +    return true;
> +  MaskLZ -= ScaleDown;
>
>    // The final check is to ensure that any masked out high bits of X are
>    // already known to be zero. Otherwise, the mask has a semantic impact
>
> Added: llvm/trunk/test/CodeGen/X86/pr33844.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/pr33844.ll?rev=308503&view=auto
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/pr33844.ll (added)
> +++ llvm/trunk/test/CodeGen/X86/pr33844.ll Wed Jul 19 11:09:46 2017
> @@ -0,0 +1,38 @@
> +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
> +; RUN: llc -o - %s | FileCheck %s
> +
> +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> +target triple = "x86_64-unknown-linux-gnu"
> +
> + at global = external global i32
> + at global.1 = external global i64
> +
> +define void @patatino() {
> +; CHECK-LABEL: patatino:
> +; CHECK:       # BB#0: # %bb
> +; CHECK-NEXT:    movl {{.*}}(%rip), %eax
> +; CHECK-NEXT:    movl %eax, %ecx
> +; CHECK-NEXT:    shrl $31, %ecx
> +; CHECK-NEXT:    addl $2147483647, %ecx # imm = 0x7FFFFFFF
> +; CHECK-NEXT:    shrl $31, %ecx
> +; CHECK-NEXT:    andl $62, %ecx
> +; CHECK-NEXT:    andl $-536870912, %eax # imm = 0xE0000000
> +; CHECK-NEXT:    orl %ecx, %eax
> +; CHECK-NEXT:    movl %eax, {{.*}}(%rip)
> +; CHECK-NEXT:    retq
> +bb:
> +  %tmp = load i32, i32* @global
> +  %tmp1 = lshr i32 %tmp, 31
> +  %tmp2 = add nuw nsw i32 %tmp1, 2147483647
> +  %tmp3 = load i64, i64* @global.1
> +  %tmp4 = shl i64 %tmp3, 23
> +  %tmp5 = add nsw i64 %tmp4, 8388639
> +  %tmp6 = trunc i64 %tmp5 to i32
> +  %tmp7 = lshr i32 %tmp2, %tmp6
> +  %tmp8 = load i32, i32* @global
> +  %tmp9 = and i32 %tmp7, 62
> +  %tmp10 = and i32 %tmp8, -536870912
> +  %tmp11 = or i32 %tmp9, %tmp10
> +  store i32 %tmp11, i32* @global
> +  ret void
> +}
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits

Hans, we might consider backporting it to 5.0 once this has baked in
the tree for a bit.
I marked https://bugs.llvm.org/show_bug.cgi?id=33844 as "blocker" just
to make sure it's in your radar.

Thanks,

-- 
Davide

"There are no solved problems; there are only problems that are more
or less solved" -- Henri Poincare